Search Images Maps Play YouTube News Gmail Drive More »
Sign in
Screen reader users: click this link for accessible mode. Accessible mode has the same essential features but works better with your reader.

Patents

  1. Advanced Patent Search
Publication numberUS7289564 B2
Publication typeGrant
Application numberUS 10/658,374
Publication dateOct 30, 2007
Filing dateSep 10, 2003
Priority dateSep 10, 2003
Fee statusLapsed
Also published asUS20050053133
Publication number10658374, 658374, US 7289564 B2, US 7289564B2, US-B2-7289564, US7289564 B2, US7289564B2
InventorsYung-Ching Chang
Original AssigneeSilicon Integrated Systems Corp.
Export CitationBiBTeX, EndNote, RefMan
External Links: USPTO, USPTO Assignment, Espacenet
Video encoding method with support for editing when scene changed
US 7289564 B2
Abstract
A video encoding method with support for editing when scene changed. The video encoding method reads and stores the pictures by the display order and detects whether the scene change occurred. The method encodes the pictures by the coding order when there are not scenes changed and encodes the pictures by a special coding process when there are scenes changed. Because the video encoding method encodes the pictures with considering the states of scenes changed and generates a new GOP when a scene change occurred, the video sequence can be cut into two parts by an image editing process without re-encoding. Therefore, the video can be edited without any loss and the editing performance of the editing process can be better.
Images(4)
Previous page
Next page
Claims(3)
1. A video encoding method with support for editing when scene changed, the distance between two reference pictures being defined as M in a GOP, the method comprising the steps of:
capturing pictures in a display order;
detecting the scene change for a picture PICn; and
coding the pictures in a coding order when there is not a scene change occurred for the picture PICn, and coding the pictures by a special processing when there is a scene change occurred for the picture PICn;
the special processing comprising:
executing a first and a third coding stages when the picture PICn−1 is not a reference picture; and
executing a second and the third coding stages when the picture PICn−1 is a reference picture;
wherein the first coding stage is to re-code the picture PICn−1 as a P-picture type, the second coding stage is to code the pictures of B-picture type preceding the picture PICn−1, and the third coding stage is to start a new GOP, to code a picture PICn+M−1 as a I-picture type, and to code the pictures PICn to PICn+M−2 as B-picture type with only referencing to the picture PICn+M−1.
2. The video encoding method of claim 1, wherein the first coding stage finishes coding the pictures of B-pictures type if there are pictures of B-pictures type preceding a previous reference picture.
3. The video encoding method of claim 1, wherein the first coding stage codes the pictures of B-pictures type if there are pictures of B-pictures type preceding the picture PICn−1.
Description
BACKGROUND OF THE INVENTION

1. Field of the Invention

The invention relates to a video encoding method, especially to a video encoding method with support for editing when scene changed.

2. Description of the Related Art

In MPEG (Moving Pictures Experts Group), there are three picture types: I-picture, P-picture and B-picture. I-pictures are coded without referring to other pictures. I-pictures provide the coded sequence with access points, which are the starting points for the decoding process, but are coded with only moderate compression. P-pictures are coded more efficiently using motion compensated prediction from a past I-picture or P-picture and are generally used as a reference for further prediction. B-pictures provide the highest degree of compression but require both past and future reference pictures for motion compensation. B-pictures are never used as references for prediction. The organization of the three picture types in a sequence is very flexible. The choice of the sequence is determined by the encoder and will depend on the requirements of the application.

Because the B-pictures must refer to the past and future reference pictures, the encoding process of the B-pictures is delayed until the future reference picture is coded. Therefore, the display order is different to the coding order. This is called the reordering of B-pictures.

In MPEG-1, there is a group-of-pictures (hereinafter called as GOP) structure used to enclose some pictures into a group for manipulation. A GOP contains one I-picture, some P-pictures and some B-pictures. A GOP begins with an I-picture, and ends before the next I-pictures, in the coding order. In MPEG-2, the GOP structure becomes an option.

Generally, an encoder employs a fixed GOP structure. The size of a GOP is defined as N, and the distance between two reference pictures is defined as M. FIG. 1 illustrates a GOP with N=15 and M=3.

Typically, if the input signal for the encoder is in NTSC (National Television System Committee) format (29.97 fps), the GOP structure with N=15 and M=3 is used. If the input signal is in PAL (25 fps) or film format (24 fps), the GOP structure with N=12 and M=3 is used. These fixed default settings can achieve a good balance between the complexity of an encoder and the coding performance of most types of videos.

Typically, the editing process would cut the whole video sequence into pieces based on the scene, and then rearrange them to form a new video sequence. If a video sequence is coded with a fixed pattern composed with only I- and P-pictures, like IPPPPIPPPP . . . , the situation is pretty simple. If a scene change occurs in an I-picture of the video sequence (IPPPPIPPPP . . . ), the video sequence can be cut into two parts without any loss. If a scene change occurs in a P-picture of the video sequence, the former part of the video sequence is in a normal operation, but the remaining part of the video sequence has to be re-encoded. The first P-picture of the of the remaining part of the video sequence has to be decoded and then re-encode to an I-picture. However, because the re-encoded I-picture differs from the original P-picture, there will be some error propagations. Re-encode the whole remaining part of the GOP until the next I-picture would be a better solution, but we would remind that re-encoding degrades the image quality significantly.

If there are B-pictures in the coded sequence, video editing becomes more complex. Please refer to FIG. 2. If a scene change occurs in the picture just after the I-picture in the coding order, like the picture B4, cutting from picture I6 can separate the two scenes easily. However, even the picture P3 and picture B4 are belong to different scenes, there would be some macroblocks in picture B4 and B5 which needs to refer to the picture P3. Therefore the picture B4 and B5 have to be re-encoded according to the picture I6 merely. Discarding the pictures B4 and B5 is the easiest way, but losing the beginning pictures of a scene would not be acceptable.

If a scene change occurs in the picture B5, the former part and the remaining part of the GOP have some pictures to be re-encoded. The picture B4 has to be re-encoded to a P-picture and then append to the former part. In the remaining part, the coded data of the picture B4 is removed and the picture B5 has to be re-encoded.

If a scene change occurs in the picture I6, the remaining part of the GOP has only to remove the coded data of the pictures B4 and B5. However, the former part of the GOP requires a complicate process. One solution is to re-encode the picture B5 to a P-picture, and then re-encode the picture B4 according to the pictures P3 and B5. Another solution is to change the two B-pictures B4 and B5 to two P-pictures.

If a scene change occurs in the picture B7, the former part of the GOP doesn't need any additional process, and a new I-picture has to be generated for the remaining part. A choice is to change the picture B7 to an I-picture, and then re-encode the remaining GOP. However, because the B-pictures usually coded with a lower quality than the I- and P-pictures, a better choice is to change the picture P9 to an I-picture, and re-encode the remaining GOP. The pictures B4 and B5 are B-pictures with only backward reference. This method can also reduce the number of P-pictures to reduce the error caused by referring to a re-encoded picture.

If a scene change occurs in the picture B8, the former part of the GOP has only to re-encode the picture B7 to a P-picture. The remaining part of the GOP can change the picture P9 to an I-picture and then re-encode the remaining part of the GOP.

Finally, if a scene change occurs in the picture P9, the former part of the GOP is processed like the situation of picture I6. For the remaining part of the GOP, the picture P9 has to be changed to an I-picture, and then re-encode the remaining GOP.

Therefore, all the other situations can be processed like the methods described above, even if the number of B-pictures between two reference pictures increases to three or more.

Generally, the I-pictures are designed for the purpose of random access and preventing of error propagation. The P-pictures use the motion compensation to remove the temporal redundancy between the current picture and the reference picture to improve the compression performance. However, if there is almost no temporal redundancy between the current picture and the reference picture, such as a scene change, coding a picture as a P-picture can't obtain any benefit. In this case, coding a picture as an I-picture can achieve the same coding quality with fewer bits. Therefore, an encoder has to detect the existence of a scene change and then start a new GOP. There are many researches of the scene change detection and the algorithm of adjusting the rate control. A general idea is to detect the difference of the current picture and the reference picture from the result of motion estimation. If more than a percentage of macroblocks select the intra-coded mode, the encoder can decide that there is only few temporal redundancy existed, and therefore a scene change can be detected.

However, when a scene change is detected, if the encoder just starts a new GOP without any effort, the re-encoding of some pictures would be unavoidable during the video sequence editing process as described above.

SUMMARY OF THE INVENTION

In view of the above-mentioned problems, an object of the invention is to provide a video encoding method capable of editing scene changes.

To achieve the above-mentioned object, the video encoding method of the present invention encodes the pictures by the coding order when there are not scenes changed and encodes the pictures by a special coding process when there are scenes changed. Because the video encoding method encodes the pictures with considering the states of scenes changed and generates a new GOP when a scene change occurred, the video sequence can be cut into two parts by an image editing process without re-encoding.

Therefore, the video can be edited without any loss and the editing performance of the editing process can be better.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 illustrates a GOP with N=15 and M=3.

FIG. 2 illustrates a video sequence with a B-picture.

FIG. 3 illustrates a GOP with a fixed structure.

FIG. 4 illustrates a first example of GOP with display order and coding order.

FIG. 5 illustrates a second example of GOP with display order and coding order.

FIG. 6 illustrates a third example of GOP with display order and coding order.

FIG. 7 shows the flowchart of the video encoding method with support for editing when scene changed of the present invention.

DETAILED DESCRIPTION OF THE INVENTION

The video encoding method with support for editing when scene changed of the invention will be described with reference to the accompanying drawings.

First, an encoder using the video encoding method with support for editing when scene changed of the present invention also has to include the scene change detecting function. The scene change detecting function has to be applied in the display order. This is because the encoder has to know where scene changes and then encodes the pictures before and after the scene change into two GOP.

Before a scene change is detected, the encoder encodes the video sequence with a fixed GOP structure. Once a scene change is detected, the encoder decides how to encode the following pictures based on the type and position in a GOP of the present coded pictures. Please note that because the B-pictures have to be coded just after the future reference picture being coded, a scene change needs to be detected before the coding happens. FIG. 2 depicts an example. The encoder captures pictures and stores them into a buffer in the display order. The picture B4 and B5 are captured and not being coded until the picture I6 is coded. If the encoder can encode a picture in each period of capturing a picture, the picture I6 is captured and encoded in the same period. In the next period, the picture B4 is encoded while the picture B7 is being captured. The picture B5 is encoded in the same period when the picture B8 is captured. The picture P9 only needs to take the picture I6 as a reference so that it can be captured and encoded in the same period.

For an encoder, which encodes the video sequence with a fixed GOP structure, the distance between two reference pictures is defined as M and a reference picture (I- or P-picture) is represented as an R. The first B-picture (in the display order) after the forward reference picture RX is called BX 1, the second B-picture is called BX 2, and so on. The final one before the back reference picture is called BX M−1. FIG. 3 illustrates an example of the GOP in the display order and coding order according to this definition.

The process methods for the scene change occurred at different location are described as following.

A. A Scene Change Occurs in the First B-picture

If there is no scene change occurred in the pictures from BA 1 to RB, the picture BA 1˜BA M−1 is captured and stored until the picture RB is captured and coded. If the scene changes in the picture BB 1, the pictures until RB would belong to the former GOP and the pictures from BB 1 would belong to a new GOP. After coding the picture BA M−1, if the encoder starts a new GOP and encodes the following pictures without referring to the picture RB, it can completely separate the video sequence into two parts. An editing process can cut the video sequence from the new GOP without any re-encoding.

There are two strategies to start a new GOP. One is to start a fixed GOP structure from I-picture. In the above example, the original picture BB 1 is changed to an I-picture RC, the following M−1 pictures are B-picture BC 1˜BC M−1, the next picture is a P-picture RD, then the following-up pictures are the M−1 B-pictures, and so on. FIG. 4 illustrates an example of this case.

However, a new GOP need not be started with an I-picture in the display order. By observing the coding order in FIG. 4, we can find that there are no B-pictures between the picture RC and RD. B-pictures can be coded with lower quality and bit rate than the I-pictures and P-pictures. If there are too many reference pictures in a short duration, the result is that each reference picture can't obtain enough bits to achieve a higher quality. Therefore, the second strategy of starting a new GOP is trying to maintain the ratio on the number of B-pictures and reference pictures. The first M−1 pictures of the new GOP are B-pictures, the next picture is an I-picture, the following pictures are M−1 B-pictures, and then the picture is a P-picture, and so on. FIG. 5 illustrates an example of this case.

It seems that the picture type of each picture is remaining the same as no scene change occurred. Actually, the difference is that the picture BB 1˜BB M−1 only refer to the back picture RC. In fact, there may not be M−1 B-pictures before the picture RC, and can be adjusted freely.

B. A Scene Change Occurs in the Second B-picture

Please refer to FIG. 3. If a scene change occurs in the picture BB 2, the picture BB 1 belongs to the former GOP and the pictures from BB 2 form a new GOP. The new GOP can be encoded with the same method described in subsection A.

A GOP can be ended by a reference picture. Therefore the picture BB 1 must be encoded as a reference picture. Further, the picture BB 1 as a P-picture but an I-picture. FIG. 6 illustrates an example of this case.

C. A Scene Change Occurs in the n-th B-picture

If a scene change occurs in the n-th B-picture after the reference picture RX, 2≦n≦M−1, the pictures until BX n−1 belong to the former GOP and the pictures from BX n form a new GOP. The new GOP is encoded with the same method described in subsection A.

Based on the method described in section B, the encoder will encode the picture BX n−1 as a P-picture, and the pictures BX 1˜BX n−2 (if any) are encoded as B-pictures by referencing to the picture RX and the new generated P-picture.

D. A Scene Change Occurs in a Reference Picture

Please refer to FIG. 3. If a scene change occurs in the picture RB, the picture BA M−1 belongs to the former GOP and the pictures from RB form a new GOP. The former GOP can be coded by the method described in subsection C. The new GOP could be encoded with the same method described in subsection A.

Finally, we can simplify an algorithm for an encoder to encode the pictures before and after a scene change separately to make the video editing of scenes without any re-encoding. FIG. 7 shows the flowchart of the video encoding method with support for editing when scene changed of the present invention.

Step 702: Capture the picture PICn in the display order and detect the scene change.

Step 704: If there is no scene change in the picture PICn, the flowchart jumps to step S706. If there is a scene change in the picture PICn, the flowchart jumps to step S708.

Step 706: Code pictures in the coding order and jump back to step S702.

Step 708: If the picture PICn−1 is not coded as a reference picture, the flowchart jumps to step S710. If the picture PICn−1 is coded as a reference picture, the flowchart jumps to step S716.

Step 710: If there are B-pictures preceding a previous reference picture, finish coding the B-pictures.

Step S712: Encode the picture PICn−1 as a P-picture.

Step S714: If there are B-pictures preceding the picture PICn−1, coding the B-pictures and jumping to step S718.

Step S716: If there are B-pictures preceding the picture PICn−1, coding the B-pictures and jumping to step S718.

Step S718: Start a new GOP and encode the picture Pn+M−1 as an I-picture.

Step S720: Encode the pictures Pn˜Pn+M−2 as B-pictures with only referencing to the picture Pn+M−1.

Because the present invention starts a new GOP when the scene change occurred, the image editor can cut the video sequence directly without re-coding. Therefore, the performance of the image editor can be increased and the image quality will not be loss.

While certain exemplary embodiments have been described and shown in the accompanying drawings, it is to be understood that such embodiments are merely illustrative of and not restrictive on the broad invention, and that this invention should not be limited to the specific construction and arrangement shown and described, since various other modifications may occur to those ordinarily skilled in the art.

Patent Citations
Cited PatentFiling datePublication dateApplicantTitle
US6038000 *Dec 23, 1997Mar 14, 2000Sarnoff CorporationInformation stream syntax for indicating the presence of a splice point
US6301428 *Dec 9, 1997Oct 9, 2001Lsi Logic CorporationCompressed video editor with transition buffer matcher
US6529555 *Aug 25, 2000Mar 4, 2003Sony United Kingdom LimitedSignal processor
US6731684 *Aug 27, 1999May 4, 2004General Instrument CorporationMethod and apparatus for detecting scene changes and adjusting picture coding type in a high definition television encoder
US6754275 *Nov 26, 1999Jun 22, 2004Matsushita Electric Industrial Co., Ltd.Decoding device and method for deciding the next decoded frame based on the frame interval
Referenced by
Citing PatentFiling datePublication dateApplicantTitle
US8194751 *Feb 13, 2004Jun 5, 2012Panasonic CorporationMoving picture coding method and moving picture decoding method
US20040161033 *Feb 13, 2004Aug 19, 2004Youji NotoyaMoving picture coding method and moving picture decoding method
Classifications
U.S. Classification375/240.13, 375/240.26, 375/E07.165, 375/E07.148, 375/E07.211
International ClassificationH04N7/12, H04N7/26, H04N7/18, H04N7/50
Cooperative ClassificationH04N19/107, H04N19/142, H04N19/61
European ClassificationH04N7/26A6C6, H04N7/26A4C2, H04N7/50
Legal Events
DateCodeEventDescription
Sep 10, 2003ASAssignment
Owner name: SILICON INTEGRATED SYSTEMS CORP., TAIWAN
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:CHANG, YUNG-CHING;REEL/FRAME:014480/0232
Effective date: 20030903
Apr 6, 2011FPAYFee payment
Year of fee payment: 4
Oct 7, 2012ASAssignment
Owner name: XIAHOU HOLDINGS, LLC, DELAWARE
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:SILICON INTEGRATED SYSTEMS CORP.;REEL/FRAME:029088/0198
Effective date: 20120926
Jun 12, 2015REMIMaintenance fee reminder mailed
Oct 30, 2015LAPSLapse for failure to pay maintenance fees
Dec 22, 2015FPExpired due to failure to pay maintenance fee
Effective date: 20151030